Intelligent monitoring of steel bridge bearings using deep learning methods

The advancement of mobile imaging technology has given rise to computer vision techniques, but there have been few attempts to use computer vision for automatic condition assessment of bridge bearings. In fact, human visual inspection still plays a major role in the assessment of bridge bearings conditions today. The objective of this research is to devise an intelligent Struc-tural Health Monitoring (SHM) system for steel bearings of bridges, with a particular focus on aged steel bearings. These components are deeply related to the load-bearing capacity and over-all structural integrity of bridges. The proposed SHM system leverages the capabilities of Deep Learning (DL), specifically Convolutional Neural Networks (CNNs), for image-based condi-tion assessment. The model is trained to perform two sequential tasks: component detection and damage classification. Initially, the model identifies the presence of bearings within the col-lected images, subsequently, it classifies the detected bearings, then it identifies a possible dam-age, providing a quantitative measure of its extension. A dataset of images showing various conditions of steel bearings and connections is collected and annotated to support the develop-ment of the SHM system.